[5/6] lirc_zilog: error out if buffer read bytes != chunk size

  Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 drivers/staging/lirc/lirc_zilog.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)
  

On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
> Signed-off-by: Jarod Wilson <jarod@redhat.com>
> ---
>  drivers/staging/lirc/lirc_zilog.c |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/staging/lirc/lirc_zilog.c b/drivers/staging/lirc/lirc_zilog.c
> index 407d4b4..5ada643 100644
> --- a/drivers/staging/lirc/lirc_zilog.c
> +++ b/drivers/staging/lirc/lirc_zilog.c
> @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep, char *outbuf, size_t n, loff_t *ppos)
>  				ret = copy_to_user((void *)outbuf+written, buf,
>  						   rbuf->chunk_size);
>  				written += rbuf->chunk_size;
> +			} else {
> +				zilog_error("Buffer read failed!\n");
> +				ret = -EIO;
> +				break;

No need to break, just let the non-0 ret value drop you out of the while
loop.

Regards,
Andy

>  			}
>  		}
>  	}


--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
  

On Wed, Mar 16, 2011 at 08:07:22PM -0400, Andy Walls wrote:
> On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
> > Signed-off-by: Jarod Wilson <jarod@redhat.com>
> > ---
> >  drivers/staging/lirc/lirc_zilog.c |    4 ++++
> >  1 files changed, 4 insertions(+), 0 deletions(-)
> > 
> > diff --git a/drivers/staging/lirc/lirc_zilog.c b/drivers/staging/lirc/lirc_zilog.c
> > index 407d4b4..5ada643 100644
> > --- a/drivers/staging/lirc/lirc_zilog.c
> > +++ b/drivers/staging/lirc/lirc_zilog.c
> > @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep, char *outbuf, size_t n, loff_t *ppos)
> >  				ret = copy_to_user((void *)outbuf+written, buf,
> >  						   rbuf->chunk_size);
> >  				written += rbuf->chunk_size;
> > +			} else {
> > +				zilog_error("Buffer read failed!\n");
> > +				ret = -EIO;
> > +				break;
> 
> No need to break, just let the non-0 ret value drop you out of the while
> loop.

Ah, indeed. I think I mindlessly copied what the tests just a few lines
above were doing without looking at the actual reason for them. I'll
remove that break from the patch here locally.
  

Jarod Wilson <jarod@redhat.com> wrote:

>On Wed, Mar 16, 2011 at 08:07:22PM -0400, Andy Walls wrote:
>> On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
>> > Signed-off-by: Jarod Wilson <jarod@redhat.com>
>> > ---
>> >  drivers/staging/lirc/lirc_zilog.c |    4 ++++
>> >  1 files changed, 4 insertions(+), 0 deletions(-)
>> > 
>> > diff --git a/drivers/staging/lirc/lirc_zilog.c
>b/drivers/staging/lirc/lirc_zilog.c
>> > index 407d4b4..5ada643 100644
>> > --- a/drivers/staging/lirc/lirc_zilog.c
>> > +++ b/drivers/staging/lirc/lirc_zilog.c
>> > @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep, char
>*outbuf, size_t n, loff_t *ppos)
>> >  				ret = copy_to_user((void *)outbuf+written, buf,
>> >  						   rbuf->chunk_size);
>> >  				written += rbuf->chunk_size;
>> > +			} else {
>> > +				zilog_error("Buffer read failed!\n");
>> > +				ret = -EIO;
>> > +				break;
>> 
>> No need to break, just let the non-0 ret value drop you out of the
>while
>> loop.
>
>Ah, indeed. I think I mindlessly copied what the tests just a few lines
>above were doing without looking at the actual reason for them. I'll
>remove that break from the patch here locally.
>
>-- 
>Jarod Wilson
>jarod@redhat.com

You might also want to take a look at that test to ensure it doesn't break blocking read() behavior.  (man 2 read). I'm swamped ATM and didn't look too hard.

It seems odd that the lirc buffer object can have data ready (the first branch of the big if() in the while() loop), and yet the read of that lirc buffer object fails.

-Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
  

On Thu, Mar 17, 2011 at 11:29:08AM -0400, Andy Walls wrote:
> Jarod Wilson <jarod@redhat.com> wrote:
> 
> >On Wed, Mar 16, 2011 at 08:07:22PM -0400, Andy Walls wrote:
> >> On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
> >> > Signed-off-by: Jarod Wilson <jarod@redhat.com>
> >> > ---
> >> >  drivers/staging/lirc/lirc_zilog.c |    4 ++++
> >> >  1 files changed, 4 insertions(+), 0 deletions(-)
> >> > 
> >> > diff --git a/drivers/staging/lirc/lirc_zilog.c
> >b/drivers/staging/lirc/lirc_zilog.c
> >> > index 407d4b4..5ada643 100644
> >> > --- a/drivers/staging/lirc/lirc_zilog.c
> >> > +++ b/drivers/staging/lirc/lirc_zilog.c
> >> > @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep, char
> >*outbuf, size_t n, loff_t *ppos)
> >> >  				ret = copy_to_user((void *)outbuf+written, buf,
> >> >  						   rbuf->chunk_size);
> >> >  				written += rbuf->chunk_size;
> >> > +			} else {
> >> > +				zilog_error("Buffer read failed!\n");
> >> > +				ret = -EIO;
> >> > +				break;
> >> 
> >> No need to break, just let the non-0 ret value drop you out of the
> >while
> >> loop.
> >
> >Ah, indeed. I think I mindlessly copied what the tests just a few lines
> >above were doing without looking at the actual reason for them. I'll
> >remove that break from the patch here locally.
> >
> >-- 
> >Jarod Wilson
> >jarod@redhat.com
> 
> You might also want to take a look at that test to ensure it doesn't break blocking read() behavior.  (man 2 read). I'm swamped ATM and didn't look too hard.
> 
> It seems odd that the lirc buffer object can have data ready (the first branch of the big if() in the while() loop), and yet the read of that lirc buffer object fails.

Generally, it shouldn't, but lirc_buffer_read uses kfifo underneath, and
in the pre-2.6.33 kfifo implementation, the retval from lirc_buffer_read
(as backported by way of media_build) is always 0, which is of course not
equal to chunk_size. So I think that in current kernels, this should never
trigger, and its partially just a note-to-self that this check will go
sideways when running on an older kernel, but not a bad check to have if
something really does go wrong.
  

Jarod Wilson <jarod@redhat.com> wrote:

>On Thu, Mar 17, 2011 at 11:29:08AM -0400, Andy Walls wrote:
>> Jarod Wilson <jarod@redhat.com> wrote:
>> 
>> >On Wed, Mar 16, 2011 at 08:07:22PM -0400, Andy Walls wrote:
>> >> On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
>> >> > Signed-off-by: Jarod Wilson <jarod@redhat.com>
>> >> > ---
>> >> >  drivers/staging/lirc/lirc_zilog.c |    4 ++++
>> >> >  1 files changed, 4 insertions(+), 0 deletions(-)
>> >> > 
>> >> > diff --git a/drivers/staging/lirc/lirc_zilog.c
>> >b/drivers/staging/lirc/lirc_zilog.c
>> >> > index 407d4b4..5ada643 100644
>> >> > --- a/drivers/staging/lirc/lirc_zilog.c
>> >> > +++ b/drivers/staging/lirc/lirc_zilog.c
>> >> > @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep,
>char
>> >*outbuf, size_t n, loff_t *ppos)
>> >> >  				ret = copy_to_user((void *)outbuf+written, buf,
>> >> >  						   rbuf->chunk_size);
>> >> >  				written += rbuf->chunk_size;
>> >> > +			} else {
>> >> > +				zilog_error("Buffer read failed!\n");
>> >> > +				ret = -EIO;
>> >> > +				break;
>> >> 
>> >> No need to break, just let the non-0 ret value drop you out of the
>> >while
>> >> loop.
>> >
>> >Ah, indeed. I think I mindlessly copied what the tests just a few
>lines
>> >above were doing without looking at the actual reason for them. I'll
>> >remove that break from the patch here locally.
>> >
>> >-- 
>> >Jarod Wilson
>> >jarod@redhat.com
>> 
>> You might also want to take a look at that test to ensure it doesn't
>break blocking read() behavior.  (man 2 read). I'm swamped ATM and
>didn't look too hard.
>> 
>> It seems odd that the lirc buffer object can have data ready (the
>first branch of the big if() in the while() loop), and yet the read of
>that lirc buffer object fails.
>
>Generally, it shouldn't, but lirc_buffer_read uses kfifo underneath,
>and
>in the pre-2.6.33 kfifo implementation, the retval from
>lirc_buffer_read
>(as backported by way of media_build) is always 0, which is of course
>not
>equal to chunk_size. So I think that in current kernels, this should
>never
>trigger, and its partially just a note-to-self that this check will go
>sideways when running on an older kernel, but not a bad check to have
>if
>something really does go wrong.
>
>-- 
>Jarod Wilson
>jarod@redhat.com

But the orignal intent of the check I put in was to avoid passing partial/junk data to userspace, and go around again to see if good data could be provided.  

Your check bails when good data that might be sitting there still.  That doesn't seem like a good trade for supporting backward compat for old kernels.
  
-Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
  

On Thu, Mar 17, 2011 at 12:16:31PM -0400, Andy Walls wrote:
> Jarod Wilson <jarod@redhat.com> wrote:
> 
> >On Thu, Mar 17, 2011 at 11:29:08AM -0400, Andy Walls wrote:
> >> Jarod Wilson <jarod@redhat.com> wrote:
> >> 
> >> >On Wed, Mar 16, 2011 at 08:07:22PM -0400, Andy Walls wrote:
> >> >> On Wed, 2011-03-16 at 16:24 -0400, Jarod Wilson wrote:
> >> >> > Signed-off-by: Jarod Wilson <jarod@redhat.com>
> >> >> > ---
> >> >> >  drivers/staging/lirc/lirc_zilog.c |    4 ++++
> >> >> >  1 files changed, 4 insertions(+), 0 deletions(-)
> >> >> > 
> >> >> > diff --git a/drivers/staging/lirc/lirc_zilog.c
> >> >b/drivers/staging/lirc/lirc_zilog.c
> >> >> > index 407d4b4..5ada643 100644
> >> >> > --- a/drivers/staging/lirc/lirc_zilog.c
> >> >> > +++ b/drivers/staging/lirc/lirc_zilog.c
> >> >> > @@ -950,6 +950,10 @@ static ssize_t read(struct file *filep,
> >char
> >> >*outbuf, size_t n, loff_t *ppos)
> >> >> >  				ret = copy_to_user((void *)outbuf+written, buf,
> >> >> >  						   rbuf->chunk_size);
> >> >> >  				written += rbuf->chunk_size;
> >> >> > +			} else {
> >> >> > +				zilog_error("Buffer read failed!\n");
> >> >> > +				ret = -EIO;
> >> >> > +				break;
> >> >> 
> >> >> No need to break, just let the non-0 ret value drop you out of the
> >> >while
> >> >> loop.
> >> >
> >> >Ah, indeed. I think I mindlessly copied what the tests just a few
> >lines
> >> >above were doing without looking at the actual reason for them. I'll
> >> >remove that break from the patch here locally.
> >> >
> >> >-- 
> >> >Jarod Wilson
> >> >jarod@redhat.com
> >> 
> >> You might also want to take a look at that test to ensure it doesn't
> >break blocking read() behavior.  (man 2 read). I'm swamped ATM and
> >didn't look too hard.
> >> 
> >> It seems odd that the lirc buffer object can have data ready (the
> >first branch of the big if() in the while() loop), and yet the read of
> >that lirc buffer object fails.
> >
> >Generally, it shouldn't, but lirc_buffer_read uses kfifo underneath,
> >and
> >in the pre-2.6.33 kfifo implementation, the retval from
> >lirc_buffer_read
> >(as backported by way of media_build) is always 0, which is of course
> >not
> >equal to chunk_size. So I think that in current kernels, this should
> >never
> >trigger, and its partially just a note-to-self that this check will go
> >sideways when running on an older kernel, but not a bad check to have
> >if
> >something really does go wrong.
> 
> But the orignal intent of the check I put in was to avoid passing partial/junk data to userspace, and go around again to see if good data could be provided.  
> 
> Your check bails when good data that might be sitting there still.  That doesn't seem like a good trade for supporting backward compat for old kernels.

Ah. Another thing I neglected to notice then. :)

Perhaps there should be a retry count check as well then, as otherwise,
its possible to get stuck in that loop forever (which is what was
happening on older kernels). Its conceivable that similar could happen on
a newer kernel for some reason.
  

On Thu, 2011-03-17 at 15:08 -0400, Jarod Wilson wrote:
> On Thu, Mar 17, 2011 at 12:16:31PM -0400, Andy Walls wrote:
> > Jarod Wilson <jarod@redhat.com> wrote:
> .
> > 
> > But the orignal intent of the check I put in was to avoid passing
> partial/junk data to userspace, and go around again to see if good
> data could be provided.  
> > 
> > Your check bails when good data that might be sitting there still.
> That doesn't seem like a good trade for supporting backward compat for
> old kernels.
> 
> Ah. Another thing I neglected to notice then. :)
> 
> Perhaps there should be a retry count check as well then, as otherwise,
> its possible to get stuck in that loop forever (which is what was
> happening on older kernels). Its conceivable that similar could happen on
> a newer kernel for some reason.

Well, lets see,

>From the perspective of userspace & lircd:

1. A specification compliance failure for a corner case isn't too bad
(bailing out on junk and leaving good data behind)

2. An unrecoverable failure for any case is very bad (spinning/hanging
on a result that won't change)

3. Sending unitialized bytes out to userspace with copy_to_user() is
very bad.
(I recall the old code would do the copy to user and always tell
userspace it got a code whether it read anything out of the buffer or
not.  IIRC, that leaked information off the stack.)


If the code as patched avoids the two very bad things (#2 and #3), then
the patch is OK by me.

Regards,
Andy

--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
  

On Thu, Mar 17, 2011 at 08:50:33PM -0400, Andy Walls wrote:
> On Thu, 2011-03-17 at 15:08 -0400, Jarod Wilson wrote:
> > On Thu, Mar 17, 2011 at 12:16:31PM -0400, Andy Walls wrote:
> > > Jarod Wilson <jarod@redhat.com> wrote:
> > .
> > > 
> > > But the orignal intent of the check I put in was to avoid passing
> > partial/junk data to userspace, and go around again to see if good
> > data could be provided.  
> > > 
> > > Your check bails when good data that might be sitting there still.
> > That doesn't seem like a good trade for supporting backward compat for
> > old kernels.
> > 
> > Ah. Another thing I neglected to notice then. :)
> > 
> > Perhaps there should be a retry count check as well then, as otherwise,
> > its possible to get stuck in that loop forever (which is what was
> > happening on older kernels). Its conceivable that similar could happen on
> > a newer kernel for some reason.
> 
> Well, lets see,
> 
> >From the perspective of userspace & lircd:
> 
> 1. A specification compliance failure for a corner case isn't too bad
> (bailing out on junk and leaving good data behind)
> 
> 2. An unrecoverable failure for any case is very bad (spinning/hanging
> on a result that won't change)
> 
> 3. Sending unitialized bytes out to userspace with copy_to_user() is
> very bad.
> (I recall the old code would do the copy to user and always tell
> userspace it got a code whether it read anything out of the buffer or
> not.  IIRC, that leaked information off the stack.)
> 
> 
> If the code as patched avoids the two very bad things (#2 and #3), then
> the patch is OK by me.

I *think* what I've got now should address both 2 and 3, with a very
minimal risk of leaving data behind, since it'll retry a couple of times
before bailing out of the loop, so it should be pretty unlikely we'd leave
any good data behind. But even if we do, like you said, this is just an IR
signal, the user can press the button on the remote again. :)